A bundle of independent, agent-neutral skills for general knowledge work. Each skill can be invoked directly; the bundle has no orchestrator, required stage sequence, state machine, checkpoint protocol, or cross-skill routing contract.
| Skill | Purpose |
|---|---|
probing |
Explore uncertain goals and competing directions with the user. |
challenge |
Pressure-test an existing artifact, plan, idea, conclusion, or decision. |
increment |
Confirm requirements, plan verifiable increments, and optionally execute them. |
align |
Keep project documentation aligned after knowledge, implementation, decision, or evidence changes. |
vnv |
Verify and validate results from the perspectives relevant to the task. |
deep-research |
Derive answers from sources, tracing, comparison, and synthesis. |
design-system |
Build, extract, or rebuild a design system and deliver it as a showcase plus a token-complete specification. |
distillation |
Run a retrospective and decide whether reusable skill guidance should change. |
restraint |
Act with restraint — do what the user wants, nothing more; avoid over-engineering. |
prune and Rationale |
Subtract what is unnecessary — obsolete workspace clutter and overgrowth in a drafted design or plan — keeping the smallest necessary and sufficient set. |
worklog |
Record traceable workspace changes and maintain current work state. |
style-calibration is maintained in the utility bundle because it is an optional writing utility rather than part of this bundle.
Before publishing:
for f in drafting/skills/*/SKILL.md; do
test "$(wc -l < "$f")" -le 100 || echo "$f too long"
rg -n '[一-龥]' "$f" && echo "$f contains Chinese characters"
done
Also verify that plugin JSON parses, internal references exist, and .agents/skills symlinks resolve.